Low-resource Hardware Research Articles

Machine learning assisted classification between diabetic polyneuropathy and healthy subjects using plantar pressure and temperature data: a feasibility study

Automated and early detection of diabetics with polyneuropathy in an ambulatory health monitoring setup may reduce the major risk factors for diabetic patients. Increased and localized plantar pressure associated with impaired pain and temperature is a combination of developing foot ulcers in subjects with polyneuropathy. Although many interesting research works have been reported in this area, most of them emphasize on signal acquisition process and plantar pressure distribution in the foot region. In this work, a machine learning assisted low complexity technique was developed using plantar pressure and temperature signals which will classify between diabetic polyneuropathy and healthy subjects. Principal component analysis (PCA) and maximum relevance minimum redundancy (mRMR) methods were used for feature extraction and selection respectively followed by k-NN classifier for binary classification. The proposed technique was evaluated with 100 min of publicly available annotated data from 43 subjects and provides blind test accuracy, sensitivity, precision, F1-score, and area under curve (AUC) of 99.58%, 99.50%, 99.44%, 99.47% and 99.56% respectively. A low resource hardware implementation in ARM v6 controller required an average memory usage of 81.2 kB and latency of 1.31 s to process 9 s pressure and temperature data collected from 16 sensor channels for each of the foot region.

Machine Learning Assisted Hit Prioritization for High Throughput Screening in Drug Discovery.

Efficient prioritization of bioactive compounds from high throughput screening campaigns is a fundamental challenge for accelerating drug development efforts. In this study, we present the first data-driven approach to simultaneously detect assay interferents and prioritize true bioactive compounds. By analyzing the learning dynamics during training of a gradient boosting model on noisy high throughput screening data using a novel formulation of sample influence, we are able to distinguish between compounds exhibiting the desired biological response and those producing assay artifacts. Therefore, our method enables false positive and true positive detection without relying on prior screens or assay interference mechanisms, making it applicable to any high throughput screening campaign. We demonstrate that our approach consistently excludes assay interferents with different mechanisms and prioritizes biologically relevant compounds more efficiently than all tested baselines, including a retrospective case study simulating its use in a real drug discovery campaign. Finally, our tool is extremely computationally efficient, requiring less than 30 s per assay on low-resource hardware. As such, our findings show that our method is an ideal addition to existing false positive detection tools and can be used to guide further pharmacological optimization after high throughput screening campaigns.

A system-on-chip solution for deep learning-based automatic fetal biometric measurement

Ultrasound assessment of fetal biometry is widely used for gestational age prediction and abnormal fetal growth diagnosis. However, manual biometric measurements require time-consuming procedures and current automated measurement approaches remain challenging in terms of accuracy and computational complexity. This paper proposes deep learning-based efficient automatic fetal biometry measurement method for system-on-chip (SoC) solution. For end-to-end automated measurements, a hardware-friendly bilateral segmentation network (H-BiSeNet) was designed and optimized for multiple fetal objects (i.e., head, abdomen and femur), and fetal biometric parameters were then calculated with a robust multiple keypoint detection algorithm. For hardware implementation with a SoC, a deep learning processing unit (DPU) was employed to accelerate the proposed segmentation network, and biometric measurements with image pre- and postprocessing were conducted on an application processing unit (APU). A total of 6000 fetal ultrasound images in the three regions were collected from 2568 subjects, and the dataset was utilized to train the proposed network. In the performance evaluation, the proposed segmentation network with optimal quantization (H-BiSeNet-Q) outperformed other segmentation networks (i.e., U-Net, U-Net++, Attention U-Net, U-Net3+, DeepLabv3 + and BiSeNetv2) in terms of the average Dice coefficient, and automated measurements with H-BiSeNet-Q were highly correlated with manual measurements (r ≥ 0.983, p < 0.01 for all parameters, i.e., biparietal diameter, head circumference, abdominal circumference and femur length). In the execution time analysis, H-BiSeNet-Q ran at 4.25 frames per second (FPS) with parallelism corresponding to 80% resource utilization in the SoC. These results showed feasibility in low-resource hardware settings such as portable ultrasound systems.

Training CNN-based Model on Low Resource Hardware and Small Dataset for Early Prediction of Melanoma from Skin Lesion Images

Melanoma is a kind of rare skin cancer that can spread quickly to the other skin layers and the organs beneath. Melanoma is known to be curable only if it is diagnosed at an early stage. This poses a challenge for accurate prediction to cut the number of deaths caused by melanoma. Deep learning methods have recently shown promising performance in classifying images accurately. However, it requires a lot of samples to generalize well, while the number of melanoma sample images is limited. To solve this issue, transfer learning has widely adapted to transfer the knowledge of the pretrained model to another domain or new dataset which has lesser samples or different tasks. This study is aimed to find which method is better to achieve this for early melanoma prediction from skin lesion images. We investigated three pretrained and one non-pretrained image classification models. Specifically, we choose the pretrained models which are efficient to train on small training sample and low hardware resource. The result shows that using limited sample images and low hardware resource, pretrained image models yield better overall accuracy and recall compared to the non-pretrained model. This suggests that pretrained models are more suitable in this task with constrained data and hardware resource.

Modified FMCW Scheme for Improved Ultrasonic Positioning and Ranging of Unmanned Ground Vehicles at Distances < 50 mm

Unmanned ground vehicles (UGVs) find extensive use in various applications, including that within industrial environments. Efforts have been made to develop cheap, portable, and light-ranging/positioning systems to accurately locate their absolute/relative position and to automatically avoid potential obstacles and/or collisions with other drones. To this aim, a promising solution is the use of ultrasonic systems, which can be set up on UGVs and can potentially output a precise reconstruction of the drone's surroundings. In this framework, a so-called frequency-modulated continuous wave (FMCW) scheme is widely employed as a distance estimator. However, this technique suffers from low repeatability and accuracy at ranges of less than 50 mm when used in combination with low-resource hardware and commercial narrowband transducers, which is a distance range of the utmost importance to avoid potential collisions and/or imaging UGV surroundings. We hereby propose a modified FMCW-based scheme using an ad hoc time-shift of the reference signal. This was shown to improve performance at ranges below 50 mm while leaving the signal unaltered at greater distances. The capabilities of the modified FMCW were evaluated numerically and experimentally. A dramatic enhancement in performance was found for the proposed FMCW with respect to its standard counterpart, which is very close to that of the correlation approach. This work paves the way for the future use of FMCWs in applications requiring high precision.

Reliability provisioning for Fog Nodes in Smart Farming IoT-Fog-Cloud continuum

Reliability is essential in Smart Farming supported by the IoT-Fog-Cloud continuum. Smart Farms’ unprotection may cause significant economic losses and low yields of production. This paper introduces an optimization model for providing reliability and, consequently, service continuity to the IoT-Fog-Cloud continuum-based smart farms. The proposed model allows Smart Farming stakeholders to find the optimal number of Fog Nodes needed to deploy farming services considering the heterogeneity in the fog capabilities, resource demands, redundancy techniques, and reliability requirements. The model was solved using linear programming and evaluated with different demands and protection schemes. Results show that protection schemes guarantee high reliability and reveal that a shared redundancy scheme reduces deployment cost and yet provides reliability. Results also indicate that deployment costs and resources depend on the type of fog-based smart farm services to serve. Moreover, they show that deploying more low-resource hardware can be less expensive for low-reliability demands than deploying with a few high-resource hardware.

Implementation of DNNs on IoT devices

Driven by the recent growth in the fields of internet of things (IoT) and deep neural networks (DNNs), DNN-powered IoT devices are expected to transform a variety of industrial applications. DNNs, however, involve many parameters and operations to process the data generated by IoT devices. This results in high data-processing latency and energy consumption. New approaches are thus being souhgt to tackle these issues and deploy real-time DNNs into resource-limited IoT devices. This paper presents a comprehensive review on hardware-and-software-co-design approaches developed to implement DNNs on low-resource hardware platforms. These approaches explore the trade-off between energy consumption, speed, classification accuracy, and model size. First, an overview of DNNs is given. Next, available tools for implementing DNNs on low-resource hardware platforms are described. Then, the memory hierarchy designs together with dataflow mapping strategies are presented. Furthermore, various model optimization approaches, including pruning and quantization, are discussed. In addition, case studies are given to demonstrate the feasibility of implementing DNNs for IoT applications. Finally, detailed discussions, research gaps, and future directions are provided. The presented review can guide the design and implementation of the next generation of hardware and software solutions for real-world IoT applications.

Survey of SLAM in Low-Resourced Hardware

Many of researches in SLAM are targeting desktops or laptop computers. Mounted in a robot platform such as Pioneer, these high computational power hardware do all the processing in SLAM. Still others, SLAM algorithms exploit GPU power to provide deep details in map reconstruction. Yet, it is desirable to deploy SLAM in a small robot without advantages from high computational power hardware. Single board computer with limited power supply and low computational power is frequently the main board available in a small robot. Therefore, it is important to consider the design solution of SLAM that targets such a system. With this in mind, current work presents a survey paper of SLAM in low-resource hardware. The main question to be answered with this current work is "How researchers deal with hardware limitation when implementing SLAM?" Classification based on a method to tackle the problem is presented as the conclusion of this paper.

Low-resource hardware implementation of the hyperbolic tangent for artificial neural networks

Artificial neural networks are a widespread tool with application in a variety of areas ranging from the social sciences to engineering. Many of these applications have reached a hardware implementation phase and have been documented in scientific papers. Unfortunately, most of the implementations have a simplified hyperbolic tangent replacement which has been the most common problem, as well as the most resource-consuming block in terms of hardware. This paper proposes a low-resource hardware implementation of the hyperbolic tangent, by using the simplest solution in order to obtain the lowest error possible thus far with a set of 25 polynomials of third order, obtained with Chebyshev interpolations. The results obtained show that the solution proposed holds a low error while simultaneously promising the use of low resources, as only third-order polynomials are used.

A Novel Structure with Dynamic Operation Mode for Symmetric-Key Block Ciphers

Modern Internet protocols support several modes of operation in encryption tasks for data confidentiality to keep up with varied environments and provide the various choices, such as multi-mode IPSec support. To begin with we will provide a brief background on the modes of operation for symmetric-key block ciphers. Different block cipher modes of operation have distinct characteristics. For example, the cipher block chaining (CBC) mode is suitable for operating environments that require self-synchronizing capabilities, and the output feedback (OFB) mode requires encryption modules only. When using symmetric-key block cipher algorithms such as the Advanced Encryption Standard (AES), users performing information encryption often encounter difficulties selecting a suitable mode of operation. This paper describes a structure for analyzing the block operation mode combination. This unified operation structure (UOS) combines existing common and popular block modes of operation. UOS does multi-mode of operation with most existing popular symmetric-key block ciphers and do not only consist of encryption mode such as electronic codebook (ECB) mode, cipher block chaining (CBC) mode, cipher feedback (CFB) mode and output feedback (OFB) mode, that provides confidentiality but also message authentication mode such as the cipher block chaining message authentication code (CBC-MAC) in cryptography. In Cloud Computing, information exchange frequently via the Internet and on-demand. This research provides an overview and information useful for approaching low-resource hardware implementation, which is proper to ubiquitous computing devices such as a sensor mote or an RFID tag. The use of the method is discussed and an example is given. This provides a common solution for multimode and this is very suitable for ubiquitous computing with several resources and environments. This study indicates a more effectively organized structure for symmetric-key block ciphers to improve their application scenarios. We can get that it is flexible in modern communication applications.

On Quality of Experience in Remote Visualization on Mobile Devices

Quality of Experience (QoE) is a relatively new concept which represents a way of measuring user satisfaction in the use of a certain kind of service. This work investigates issues related to the QoE in manipulating 3D scenes on mobile devices, by focusing on scenarios based on the remote visualization paradigm where a remote server is in charge of computing a flow of compressed images to be delivered to client devices. A novel approach able to dynamically set the encoding parameters at the server side is presented; the considered parameters are frame resolution, frame rate and image quality. The proposed solution is able to tune the above parameters according to both user preferences and network performance. Experimental tests are exploited to assess the relationship between the involved parameters and the QoE. Results obtained by considering low resource hardware (e.g. mobile devices) and unreliable connections (e.g. wireless networks) are presented. User feedback proves the effectiveness of the proposed approach.